MiniMax H3 Max · Video Extend

Some shots need a beat more breathing room and you don't find that out until you watch the render. Extend is the fix: hand over the clip, say what happens next, get your clip back longer. The source footage isn't billed, so you're paying for the new seconds only - and on a 15-second maximum that arithmetic stays reasonable.

What it is

The MiniMax H3 Max continuation node in the PopNow pack. Cloud-only, from about 7.5 credits ($0.075) per second of new footage, 5 to 15 seconds of extension at 480p or 768p. H3 Max is the fast H3 variant with native audio - the ambience and effects of the extension are generated with the frames, so the new segment doesn't arrive silent next to a source that had sound.

How it works

The node converts your clip to MP4, uploads it, and requests a segment that begins on the source's last frame. What comes back is a single file: your original with the new footage joined on. That's the whole mechanism, and it's why this beats the manual version - you never touch a timeline, never hunt for the exact cut point, never try to match grain by hand.

Worth knowing: uploads are cached by content hash for 12 hours, which means iterating on the prompt is cheap-ish (each render still costs, the upload doesn't). And the node holds the API's quoted maximum before rendering starts, so a badly-priced queue doesn't silently empty the wallet.

The inputs

• video_url - the clip to continue. MP4, MOV or WebM, up to 60 seconds. It's a socket, so come in from a video loader, and note the upload ceiling is 90 MiB - a heavy 1080p file gets rejected before any generation happens.
• prompt - next action, camera move and sound. All three, not just the action.
• duration_seconds - 5 to 15, default 8. The floor is 5 on this node; if you want a 3-second tail, use the H3 LoRA or Singularity LoRA extend nodes instead.
• resolution - 480p or 768p. Match your source. Extending a 480p clip at 768p is how you get a visible sharpness step at the join, and extending a 768p clip at 480p is how you throw away detail on the second half.
• enable_prompt_expansion - off by default; adds a flat per-request fee when on.
• last_image_url (optional) - an image the new segment should close on. Handy when the extension needs to land on a specific composition for a cut.

Outputs: video (into Save Video) and task id (API console, support). No seed - you can't re-roll a take by number, so download what you like.

Making the seam disappear

Three practical rules from doing it a few times. Match resolution exactly. Repeat the look of the source in the prompt rather than only describing new action - "same overcast light, same handheld drift, she turns toward the window" keeps the continuation in the same film. And keep the extension shorter than the source: a 4-second clip plus 12 seconds of extension reads as one shot and then a different shot.

Audio needs the same care. If the source has a specific ambience and your extension prompt describes silence, the audio seam will be more obvious than the picture seam. Name the sound you're continuing.

Install

Manager → search PopNow → Install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/popnow-ai/popnow-comfyui

Restart, add a key from popnow.ai/api/console/keys in Settings > PopNow, restart again. Zero Python dependencies - pyproject.toml declares none, and the pack runs on ComfyUI's bundled comfy_api.

Traps

The 90 MiB upload cap is the one that wastes real time. Trim first. PopNow 40004 means the model won't serve that settings combination and names the field to fix; PopNow 40201 means no credits. And cancelling in ComfyUI stops the wait, not the render - the job keeps going on PopNow, gets charged if it succeeds, and lands in the API console. If you're experimenting with expensive durations, set Max credits per run in Settings so an over-quoted run stops before anything is held.